Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

When and What to Recommend: Joint Modeling of Timing and Content for Active Sequential Recommendation

Jin Chai, Xiaoxiao Ma, Jian Yang, Jia Wu

TL;DR

This work addresses active sequential recommendation by jointly modeling the Time of Interest (ToI) and Item of Interest (IoI). It introduces PASRec, a diffusion-based framework that uses ToI predictions to guide IoI generation, improving the timing and relevance of recommendations. The method optimizes a joint objective and offers theoretical insights showing a tighter ELBO and increased mutual information between ToI and IoI. Empirical results on five datasets across two data splits show PASRec consistently outperforms eight baselines, validating the practicality of timing-aware, diffusion-based active recommendations.

Abstract

Sequential recommendation models user preferences to predict the next target item. Most existing work is passive, where the system responds only when users open the application, missing chances after closure. We investigate active recommendation, which predicts the next interaction time and actively delivers items. Two challenges: accurately estimating the Time of Interest (ToI) and generating Item of Interest (IoI) conditioned on the predicted ToI. We propose PASRec, a diffusion-based framework that aligns ToI and IoI via a joint objective. Experiments on five benchmarks show superiority over eight state-of-the-art baselines under leave-one-out and temporal splits.

When and What to Recommend: Joint Modeling of Timing and Content for Active Sequential Recommendation

TL;DR

This work addresses active sequential recommendation by jointly modeling the Time of Interest (ToI) and Item of Interest (IoI). It introduces PASRec, a diffusion-based framework that uses ToI predictions to guide IoI generation, improving the timing and relevance of recommendations. The method optimizes a joint objective and offers theoretical insights showing a tighter ELBO and increased mutual information between ToI and IoI. Empirical results on five datasets across two data splits show PASRec consistently outperforms eight baselines, validating the practicality of timing-aware, diffusion-based active recommendations.

Abstract

Sequential recommendation models user preferences to predict the next target item. Most existing work is passive, where the system responds only when users open the application, missing chances after closure. We investigate active recommendation, which predicts the next interaction time and actively delivers items. Two challenges: accurately estimating the Time of Interest (ToI) and generating Item of Interest (IoI) conditioned on the predicted ToI. We propose PASRec, a diffusion-based framework that aligns ToI and IoI via a joint objective. Experiments on five benchmarks show superiority over eight state-of-the-art baselines under leave-one-out and temporal splits.

Paper Structure

This paper contains 40 sections, 2 theorems, 21 equations, 5 figures, 7 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Works
  3. Sequential Recommendation
  4. Temporal Information for SR
  5. Preliminaries
  6. Task Definition
  7. Diffusion Models
  8. Forward Process
  9. Reverse Process
  10. Inference Process
  11. Methodology
  12. Overview of PASRec
  13. Time Encoding Functions
  14. Sinusoidal Function
  15. Gaussian Kernel Function
  16. ...and 25 more sections

Key Result

proposition 1

Minimizing the loss function of PASRec implicitly encourages the model to maximize the mutual information between the ToI and IoI representations.

Figures (5)

  • Figure 1: Two users share the same sequence of interactions, but their different interaction times result in distinct next item and time of interest.
  • Figure 2: The framework of PASRec. The left part illustrates how the model jointly trains the ToI and IoI prediction modules. The right part shows how the well-trained model infers the next item of interest based on the user’s historical interaction sequence and the corresponding real interaction times.
  • Figure 3: Overview of Time Encoding Functions.
  • Figure 4: The Impact of Incorporating ToI Information
  • Figure 5: The Accuracy Distribution of ToI Prediction

Theorems & Definitions (2)

  • proposition 1
  • proposition 2